GitTables Benchmark: Column Type Detection for Semantic Annotation

A benchmark subset of 1,101 tables from the GitTables corpus curated for evaluating column type detection systems. It was created by Madelon Hulsebos of the University of Amsterdam for the SemTab 2021 challenge's CTA task. The dataset provides ground truth annotations linking table columns to semantic types from the DBpedia and Schema.org ontologies.

Use Cases

• Benchmarking column type detection algorithms based on ontology-mapped ground truth.
• Training machine learning models for semantic annotation of table columns.
• Evaluating the performance of systems in the SemTab 2021 CTA task framework.
• Studying the relationship between raw column names and formal semantic types.

Strengths

• Contains 1,101 tables specifically curated for benchmarking.
• Provides ground truth annotations mapped to two established ontologies: DBpedia and Schema.org.
• Was featured in the SemTab 2021 challenge, indicating a recognized evaluation standard.

Limitations

• Column-level documentation is absent; original column names are anonymized to generic identifiers like 'col_0'.
• Row count and total data size are unknown, which may limit suitability assessment.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

Subset of tables from the GitTables corpus.

Collection Method

Curated for benchmarking; likely sampled and annotated by researchers.